The relative utilities of genome-wide, gene panel, and individual gene sequencing in clinical practice

The use of Southern blots to assess the clonality of antigen receptor loci to aid the diagnosis of lymphoma in the 1980s marked the beginning of the era of cancer molecular diagnostics.¹  Polymerase chain reaction²  (PCR) emerged in the 1990s, and its ease of use and versatility helped transition molecular diagnostics from a specialty service that was only available in research centers to standard practice performed in most clinical labs. Gone were the days of freezing or rapidly processing fresh samples to preserve long genomic DNA. PCR enables tests to be performed on DNA isolated from formalin-fixed, paraffin-embedded samples with <100 ng of input DNA (and, increasingly, considerably less) and completed within several hours or a few days of initiation by using automated systems with clinical-grade quality control.^3,4  Alterations in DNA or RNA present in a small fraction of cells in a heterogeneous specimen can be selectively amplified with high fidelity and identified. During the first decade of the 21st century, as our knowledge about the linkage of specific mutations with disease entities grew, so, too, did the number of individual tests designed, with the use of a broad variety of methods, to detect these alterations (henceforth, referred to as single gene assays). By 2010, it was common for a molecular diagnostic laboratory in an academic medical center to offer ≥15 single gene assays. Adding and maintaining these tests has become increasingly burdensome. The introduction of next-generation sequencing^5,6  (NGS) technology in the last decade quickly captured the attention of molecular diagnosticians. The ability to perform simultaneous analysis of scores to thousands of genetic alterations opened numerous possibilities, from testing a panel of selected targets in a multiplex fashion (henceforth, referred to as gene panel assay) to sequencing the exome (all coding exons of all known genes) or even genome (referred to as genome-wide assay herein). Medical professionals, public health administrators, and payers alike are faced now not with the question of whether testing can be done, but rather when and which tests are clinically relevant. Although in the world of “omics” knowing more seems better, finding the optimal formula to deploy limited resources and balance it with the desire for scientific and medical advancement can be a challenge. We summarize the trade-offs when considering among these testing strategies (Table 1).

The single-gene assays have been the mainstay of cancer molecular diagnostics for 3 decades. They generally involve amplification of the target of interest (DNA or RNA) from samples and normal controls followed by one of a number of analysis techniques, including direct sequencing (Sanger or pyrosequencing), size assessment (electrophoresis), restriction digestion, or hybridization with sequence-specific probes; modern single gene assays often combine amplification and analysis in a single step. The main advantages of single gene assays are that they are relatively quick to perform, inexpensive, and straightforward to develop, validate, run, and interpret. Different tests typically share relatively inexpensive equipment that is well established and familiar to use and maintain, and do not require sophisticated mathematical processes to interpret. They are highly customizable and can be tailored to the specific needs of different clinical scenarios, including the need for linear and precise quantitative assays (eg, BCR-ABL1), ultrasensitive assays (eg, KIT D816V), analysis of genomic regions that are chemically challenging for multiplexed methods (eg, CEBPA), and analysis of genomic regions that are computationally challenging for NGS methods (eg, FLT3).

The major drawback of single gene assays is their limited scope. Each individual test requires time, resources, and sample. As the number of assays rises, the feasibility of efficiently performing all of them diminishes. Eventually, the complexity of performing many individual assays exceeds the benefit of customization and the ability to deliver all of the necessary results efficiently and cost effectively. Additionally, many single gene assays have limited sensitivity, particularly methods such as Sanger sequencing, and cannot reliably detect subpopulations or work with limited samples. Moreover, developing single gene assays, although each is relatively simple, still takes considerable time and resources, and it is becoming infeasible for an individual clinical laboratory to keep deploying new assays at the same pace with which medical science has been evolving in the past 10 years. On the other hand, for scenarios where frequent monitoring of a small number of targets is needed, single-gene assays may still be the format of choice.

The availability of clinical-grade, fast, inexpensive, benchtop next generation sequencers, with prepacked analysis software and reagents, has fueled the rapid growth and spread of this disruptive technology (NGS) into clinical laboratories in recent years. NGS can simultaneously sequence millions of DNA molecules in parallel. These DNA molecules can be random, unselected fragments derived from the whole genome of an organism, or enriched for genetic regions of interest. Although many different enrichment methods have been devised, 2 target enrichment strategies have been employed for clinical applications. “Hybrid-capture” enrichment requires synthesized oligonucleotide “baits” (typically 60-100 bp oligonucleotides) specific for sequences of interest and using them in hybridization reactions to capture corresponding DNA from samples for subsequent amplification and sequencing. This design can be optimized to improve copy number quantitation and to include introns for fusion breakpoint detection. The PCR-based “amplicon” enrichment method relies on synthesized PCR primers flanking the sequences of interest and amplification of intervening sequences from samples for sequencing. As such, it requires higher-quality DNA, is subject to variation of amplification efficiency across different amplicons, and cannot detect structural alterations where no flanking primers can be designed a priori. Compared with hybrid-capture, amplicon-based library preparation is faster, costs less, and can provide better sensitivity for subclonal populations or heterogeneous samples.

Today, the majority of the hybrid-capture–based gene panel assays test for several hundred genes, mostly from coding exons with or without a smaller number of selected introns. The amplicon-based panels tend to be smaller (<100 genes) and target the coding exons of the tumor suppressor genes and selected mutation hotspots of oncogenes. NGS panels have a very low incremental cost for additional genes and, therefore, one of the benefits of these panels is their capacity to also include a large number of investigational genes with potential clinical utility at little added cost.

The limitations of gene panel assays are the upfront investment in equipment and the cost of the sequencing reagents, which makes it only financially viable to run the samples in batches and thus impractical when the overall specimen volume is too small; however, given an adequate sample volume, the cost savings of a panel may be realized when compared with relatively few single gene tests, depending on scale and methodology. Despite the wider availability of the technology in recent years, experienced personnel, both technical and professional, are still in short supply. The lack of expertise leads to uneven quality in analysis and interpretation of the complex data generated by these tests, and the results are difficult to incorporate into medical records in a structured format, making it a challenge to follow patient results over time. The regulatory and reimbursement frameworks for clinical laboratories that have been in practice for 40 years is largely inappropriate for gene panel assays, and the lack of clarity about how to validate, control, and bill for these tests has limited their deployment in hospital laboratories.

Tumor exome sequencing refers to a special form of hybrid capture gene panel assay that includes all (or nearly all, because some genes cannot be reliably captured or disambiguated from homologous regions) of the known human genes. It is technically identical to other gene panel assays except that the amount of data generated is more than an order of magnitude greater. It eliminates the need to select the content of the panel and does not need to be revised so it is easier to maintain. However, this approach increases the computational requirements, including data storage costs, and the majority of the additional genetic data provided exceeds our ability to understand and apply it. Despite the significant increase in data, because most of the commercially available bait sets do not capture introns, it cannot detect most structural variants with intron breakpoints.

Analysis of an entire genome, once a massive multicenter project that cost billions and took decades, is now feasible in 1 laboratory, in a few days, for around $1000. Because there is no enrichment step, the test is simpler to perform, although a massive amount of data is generated, and the limiting part of the procedure switches from chemistry to computation. The main advantage of genome-wide analysis is the inclusion of “everything”: fusions and copy number changes are more comprehensively assessed than in smaller panels, whereas the single nucleotide variants are still detectable. One particularly promising application is “low-pass” whole-genome analysis, which enables copy number analysis and structural variant detection from fragmented DNA that compares well with cytogenetic analysis.⁷  The main drawback of genome-wide analysis is also, however, the inclusion of “everything”: millions of interindividual variations that are unique to each human and, therefore, nigh impossible to interpret, not to mention the lack of coverage depth associated with generating so much sequence data per patient, thus limiting the ability to interrogate subclones or heterogeneous samples.

With such an intense computational requirement and low throughput, few clinical laboratories have embraced genome-wide sequence analysis outside of specialty centers, primarily working on constitutional genetics rather than cancer applications.

Myeloid neoplasms are diseases of the hematopoietic stem/progenitor cells. There are significant overlaps in the mutational profiles among different entities as currently defined in the latest World Health Organization (WHO) classification.⁸  Certain genes may be more prevalent in one diagnostic entity than others, but none is exclusive to any one entity. The JAK2 V617F mutation, for example, is found in most patients with polycythemia vera,⁹  but is also found in patients with other myeloid neoplasms.^10,11  Furthermore, patients with myeloid neoplasms commonly have mutations in >1 gene, each with independent prognostic implications. Thus, it is more practical, time-saving, and cost-effective to perform gene panel assays than multiple single gene assays for patients with suspected or newly diagnosed myeloid neoplasms. The majority of myeloid gene panel assays are amplicon based to meet the short turnaround time needed for patients with acute leukemia. Although the list of genes varies among the panels performed at different institutions, they share a core set of 20 to 25 genes that are frequently mutated in myeloid neoplasms.¹²  These include NPM1, CEBPA, and RUNX1 (to classify acute myeloid leukemia [AML] according to the 2016 WHO classification⁸  and 2017 European LeukemiaNet recommendations¹³ ); FLT3, KIT, ASXL1, WT1, and TP53 (to stratify patients into prognostic groups^14-21 ); JAK2, MPL, CSF3R and CALR (to diagnose patients with myeloproliferative neoplasms [MPNs]^9,22-26 ); IDH1 and IDH2 (to identify patients for targeted therapy^27-29 ); TET2 and DNMT3A (to evaluate clonal hematopoiesis of indeterminate potential and prognosis^30-32 ); NRAS, KRAS, and PTPN11 (to identify RAS pathway alterations³³ ); and SF3B1, SRSF2, U2AF1, ZRSR2, ASXL1, EZH2, BCOR, and STAG2 (to assess risk for myelodysplastic syndrome [MDS]^21,34-36 ). Pappaemanueil et al³⁷  recently applied a gene panel to further classify AML, identifying distinct categories with different prognoses, independent of traditional prognostic stratification, including TP53-mutated/aneuploid AML, IDH1/2-mutated AML, NPM1-mutated AML, and AML with mutations in any of a number of genes involved with splicing, chromatin, or transcription.

Beyond the initial testing at diagnosis, the value or the optimal interval of repeated testing with gene panel assays remains to be established.³⁴  However, when MDS or MPN patients progress to AML, they frequently acquire secondary alterations that can be therapeutic targets and, therefore, repeated gene panel testing may be warranted.³⁶  Panel analysis enables a richer assessment of the subclonal architecture of each patient's disease and changes in the relative abundance of clones during therapy.

The currently available gene panel assays lack the sensitivity to be used to detect minimal residual disease (MRD). In patients with AML treated with induction chemotherapy, there may still be value in ultrasensitive single gene assays to monitor MRD status with genes such as NPM1³⁸  or IDH1,³⁹  but it is difficult to generalize such an approach to genes where mutations do not occur in a single location (eg, DNMT3A³⁹ ). The KIT D816V mutation in mastocytosis also requires a single gene assay, particularly for blood testing, because of the extremely low burden of circulating cells bearing the mutation. This is most typically performed by allele-specific PCR^40,41  or an emerging technology for ultrasensitive detection, droplet digital PCR.⁴² 

Although FLT3 and CEBPA are included in many gene panel assays, they offer unique challenges for NGS. Internal tandem duplications (ITD) in FLT3,⁴³  which are associated with a poor prognosis in AML patients with a normal karyotype,^44,45  have variable boundaries and lengths, which poses a challenge for NGS assays that amplify short-read (<150 bp) sequences that are aligned and assembled with bioinformatics algorithms. Depending on the length and position of the duplication relative to the DNA fragments sequenced, a duplicated fragment may be misinterpreted by these algorithms as identical to the original and escape detection.^46,47  Novel informatics approaches may be considered to improve the sensitivity for these duplications, such as the FLT3 tandem finder algorithm published by McKerrell et al,⁴⁸  or by targeted de novo assembly of sequence reads. However, given this limitation in sensitivity, a single gene assay (ie, PCR-electrophoresis with or without Sanger sequencing) may be required to ensure adequate detection of these important alterations. Sanger sequencing is insensitive to minor populations, however, and this approach may fail to detect diversity of duplications in the same sample. CEBPA is a single-exon gene on chromosome 19 with a very high content of G:C nucleotides. This very high content of G:C nucleotides affects reaction chemistry, making it difficult to amplify properly in a multiplexed PCR reaction where conditions are optimized for sequences with average G:C content. However, this can readily be overcome in a single gene assay for which PCR conditions are tailored specifically for that gene.⁴⁹ 

Although chromosomal rearrangements can be detected in hybrid-capture–based NGS gene panel assays,^48,50  these designs involve a significant increase in the target regions to be sequenced and lack the sensitivity and linearity to be useful as a monitoring tool. RNA-based NGS approaches are promising and likely to overcome some of these limitations in the future.⁵⁰  Currently, quantitative amplification of BCR-ABL1 transcripts remains an essential single gene assay for the management of chronic myeloid leukemia (CML) patients, with the advent of targeted ABL1 inhibitors.⁵¹  The precise, linear quantitation, over 4.5 log range of concentration, needed today to monitor response to ABL1 inhibitor therapy requires a dedicated single gene assay, which is typically performed by real-time quantitative reverse transcription polymerase chain reaction (RT-PCR).⁵²  Similarly, for CML patients who progress on an ABL1 inhibitor, a single gene assay for acquired mutations in ABL1 is required to select a new inhibitor.^53,54  The presence of abundant wild-type ABL1 sequence in admixed benign cells necessitates a single gene assay capable of amplifying a long nucleic acid molecule that contains BCR and the downstream sequence in ABL1 where resistance mutations occur.

However, despite these few exceptions, the power of NGS to assess multiple genetic targets at once, in smaller samples, is transforming the practice of molecular diagnostics in hematologic malignancies as single gene tests give way to larger panels. This will likely progress further, and tumor exome or genome-wide assays may eventually be inexpensive and fast enough to perform on a routine basis at various stages of testing for myeloid neoplasms. Similarly, advances in single-cell genomic analysis, droplet digital partitioning technology, and molecular barcoding strategies will drive analytic sensitivity even lower, potentially enabling reliable MRD detection or, in the case of CML, combining MRD monitoring with the detection of secondary resistance mutations.

Unlike the myeloid neoplasms where disease entities (ie, MDS or AML) represent a spectrum of a single process that originates from hematopoietic stem/progenitor cells, lymphoid malignancies represent more distinct entities with different pathogenesis and origins.⁵⁵  Thus, although one could compile a list of genes needed for a lymphoma panel, in practice, the diagnoses rendered by morphology and immunophenotyping can typically narrow the list of relevant genes to a smaller number. In this regard, the value of the gene panel testing is more for operational simplification within the laboratory, maintaining fewer tests overall, than for clinical utility.

In addition, ascertaining the diversity of the rearranged immunoglobulin or T-cell receptor alleles in neoplastic cells is often useful in diagnosis, but has a very different design and analysis from mutation testing. Although clonality testing is still typically performed by dedicated PCR-electrophoresis assays,⁵⁶  NGS methods are also available.⁵⁷  NGS-based clonality tests offer superior resolution and depth, enabling not only the establishment of clonality, but also the monitoring of minimal residual disease⁵⁸  and, by analyzing the entire repertoire of immunoglobulin and T cell receptor loci in a specimen, the potential to reveal new relationships between the immune response and cancer. Lastly, although chromosomal rearrangements involving the immunoglobulin or T-cell receptor genes are common in lymphomas, they typically cause overexpression of the rearranged oncogenes and not a chimeric transcript; RNA-based assays aimed at the detection of fusion transcripts would be suboptimal methods, whereas DNA-based assays would need to identify the actual breakpoints that often spread across several hundred kilobases of DNA, which is impractical for most NGS assay designs short of whole-genome sequencing. NGS-based gene fusion tests can theoretically detect these, and low-pass whole-genome sequencing is particularly promising in this regard, but cytogenetic assays, either fluorescence in situ hybridization (FISH) or karyotype, remain important complementary and confirmatory assays.

For patients with B acute lymphoblastic leukemia (B-ALL), chromosomal rearrangement and karyotypic findings, including BCR-ABL1, ETV6-RUNX1, and KMT2A (MLL), are essential for establishing prognosis and therapy. With the inclusion of BCR-ABL1-like ALL as a provisional entity in the 2016 WHO classification,⁸  the number of fusions that need to be tested may exceed 200. Although these rearrangements can be tested by multiplexed FISH or RT-PCR assays, the time and resource requirements are substantial. MLL, for example, has >100 known fusion partners,⁵⁹  making RT-PCR impractical, and thus break-apart FISH probes are typically used. Although ETV6-RUNX1 and MLL fusions are typically detected by cytogenetics, BCR-ABL1 fusions could also be tested by RT-PCR. BCR-ABL1-like ALL,⁶⁰  characterized by a similar expression pattern to Philadelphia chromosome-positive (Ph⁺) ALL, but lacking the BCR-ABL1 fusion, has been described in 15% to 20% of childhood B-ALL. These patients have an adverse prognosis, and preclinical data and small case series have suggested that they may be amenable to targeted therapy,^61,62  and such targeted approaches are being evaluated in open clinical trials in pediatric ALL. More than 60 fusions have been described for BCR-ABL1-like ALL with a variety of 5′ and 3′ fusion partners. A targeted RNA sequencing assay for fusion oncoproteins that involves capture of known 3′ fusion partners and sequencing into unknown 5′ partners may be a better option. Many of these patients have an intragenic deletion of the IKZF1 gene,^63-65  which can be detected by NGS panels, but is more commonly detected by a single gene multiplexed ligation probe amplification assay.

Several other genes have prognostic and/or diagnostic significance for ALL patients, including PAX5, JAK1, JAK2, FBXW7, and IKZF1. The evidence supporting the utility of these in clinical practice is emerging rapidly, and the need to perform each of these favors a panel.

Only a handful of substitution mutations have demonstrated clinical utility in lymphoid neoplasia, although many frequently mutated genes are emerging biomarkers (Tables 2 and 3). The activating L265P mutation in MYD88 is present in <30% of many types of B-cell lymphomas, but in nearly all cases of Waldenstrom's macroglobulinemia (WM)/lymphoplasmacytic lymphoma (LPL)⁶⁶ ; the lack of an MYD88 mutation in an apparent LPL is reason to reconsider the diagnosis. This test is not only useful for the diagnosis of WM/LPL, but it is currently being evaluated as a marker for monitoring disease burden, for which an ultrasensitive test is required, necessitating a single gene assay, typically allele-specific PCR. This is particularly important for the use of peripheral blood samples in disease monitoring.

The MYD88 L265P mutation leads to activation of the NF-κB pathway via BTK, prompting treatment with the BTK inhibitor ibrutinib.⁶⁷  A subset of LPL patients who were resistant to ibrutinib treatment harbored C-terminal truncation mutations of CXCR4.⁶⁸  Other known causes for ibrutinib resistance include the acquisition of BTK C418S and PLCG2 R665W mutations.⁶⁹  Although these genes have not yet been incorporated into standard clinical practice, they are often included in NGS panels, anticipating future need.

Like MYD88, >95% of BRAF gain-of-function mutations occur in 1 hotspot, at codon V600. BRAF mutations are found in a variety of solid tumors, but are relatively uncommon in hematolymphoid neoplasms with 2 notable exceptions: hairy cell leukemias (HCLs)⁷⁰  and histiocytic neoplasms, including Langerhans cell histocytosis (LCH)⁷¹  and Erdheim–Chester disease.⁷²  Variant forms of HCL may have mutations in other genes in the same signaling pathway, with MAP2K1 most commonly reported.^73,74  Testing in these diseases is primarily used to assist in establishing diagnosis, as the absence of BRAF mutations in these diseases is grounds to reconsider the diagnosis.

Other genes have emerging clinical utility and may be included in multiple gene panels for lymphomas, although their application is often limited to atypical disease presentations and, as such, few laboratories offer these as single gene assays. For example, activating mutations in the TCF3 (E2A) transcription factor and/or loss-of-function mutations in its inhibitory partner ID3 are seen in 30% and 60% to 70% of Burkitt lymphoma cases, and are rare in other lymphoid neoplasms.^75,76  Another mutation with diagnostic utility in a specific context is RHOA G17V, which is seen in ∼70% of angioimmunoblastic T-cell lymphoma and 15% to 20% of peripheral T-cell lymphoma, not otherwise specified, but is almost never found in other T- or B-cell lymphomas or myeloid neoplasia.^77-79 

Prognostic mutations in lymphoid neoplasia include C-terminal PEST domain mutations in NOTCH1, which are seen in ∼15% of chronic lymphocytic lymphoma (CLL) patients and are associated with poor outcomes.^80,81  Mutations in SF3B1, diagnostic of RARS in patients with MDS, also have utility as adverse prognosticators in CLL, where it is seen in 15% to 20% of cases.⁸²  Loss-of-function mutations in TP53 and ATM, both involved with DNA repair, are also adverse prognostic indicators, in a variety of lymphoid malignancies.⁸³  The Eastern Cooperative Oncology Group has recently proposed a combined follicular lymphoma (FL) risk model called m7-Follicular Lymphoma International Prognostic Index (FLIPI) based on the mutation status of 7 genes (EZH2, ARID1A, MEF2B, EP300, FOXO1, CREBBP, and CARD11) and the performance status according to the FLIPI.⁸⁴  The m7-FLIPI model stratifies patients into high-risk and low-risk groups that, in a retrospective analysis, showed improved prognostication compared with FLIPI alone.

Targeted therapies are being developed in lymphoid malignancies. Along with the use of BTK inhibitors for malignancies with activation of NF-KB signaling, such as the MYD88 mutant LPL, activating STAT3 mutations are being targeted for patients with large granular lymphocyte leukemia in preclinical and early-stage trials.⁸⁵ 

The role of molecular diagnostics in the management of hematologic malignancies has expanded dramatically in the past decade. Although some of the targets that are needed for patient care today are still tested by traditional single gene assays, either because of a need for precise quantitation (BCR-ABL1 in CML) or ultrasensitive detection (KIT D816V in mastocytosis, MYD88 L265P in WM), are particularly challenging for NGS methods (FLT3 ITD, CEBPA), or because they require a unique type of assessment (IGH and TCR clonality assays), NGS-based gene panel assays have gained wide clinical acceptance for a larger number of genetic alterations that are amenable to such an approach. Genome-wide assays, although providing even more scientific information than panels, have not proven utility beyond the targeted panels and, therefore, remain investigational tools at this time. However, the history of cancer genomic testing has taught us that as technology improves and cost lowers, the trend is toward more comprehensive testing for more targets. As our capacity to interrogate the genome continues to advance, our ability to interpret and use these data has to keep pace in order to maximize the benefit to patients.

The authors thank Maria Brandao for administrative support and assistance merging contributions and editorial suggestions from the authors.

Contribution: F.C.K., B.G.M., R.C.L., and N.I.L. participated in the conception, drafting, editing, and revision of the manuscript.

Conflict-of-interest disclosure: R.C.L. is a consultant for Takeda Pharmaceuticals. The remaining authors declare no competing financial interests.

Correspondence: Neal I. Lindeman, Molecular Diagnostics, Department of Pathology, Brigham and Women's Hospital, 75 Francis St, Shapiro Cardiovascular Center 5-020, Boston, MA 02115; e-mail: nlindeman@partners.org.